Biol. Lett.
doi:10.1098/rsbl.2007.0543
Published online
Animal behaviour
Driving a hard bargain: sex
ratio and male marriage
success in a historical
US population
Thomas V. Pollet* and Daniel Nettle
Centre for Behaviour and Evolution, Newcastle University,
Newcastle upon Tyne NE2 4HH, UK
*Author and address for correspondence: Henry Wellcome Building for
Neuroecology, Newcastle University, Framlington Place, Newcastle upon
Tyne NE2 4HH, UK ([email protected]).
Evolutionary psychologists have documented a
widespread female preference for men of high
status and resources, and evidence from several
populations suggests that this preference has
real effects on marriage success. Here, we show
that in the US population of 1910, socioeconomic
status (SES) had a positive effect on men’s
chances of marrying. We also test a further
prediction from the biological markets theory,
namely that where the local sex ratio produces
an oversupply of men, women will be able to
drive a harder bargain. As the sex ratio of the
states increases, the effect of SES on marriage
success becomes stronger, indicating increased
competition between men and an increased
ability to choose on the part of women.
Keywords: evolutionary psychology; marriage;
sex ratio; biological markets; socioeconomic status
1. INTRODUCTION
Evolutionary psychologists have documented a widespread preference among women for men of high status
and resources as marriage partners (Buss & Barnes
1986; Buss 1989; Buunk et al. 2002). Since the
preferences expressed in these studies are generally
hypothetical, based on fictional vignettes, one might
question whether this expressed preference translates
into real behaviour. However, studies using data on real
marital outcomes tend to confirm that the female
preference is operative and influential on behaviour.
Men of high status and/or with high resources have
relatively increased mating and marital success in contemporary traditional societies (e.g. Kipsigis: Borgerhoff
Mulder 1990), historical European populations (e.g.
nineteenth-century Sweden: Low 1990) and contemporary developed world populations (e.g. Pérusse 1993;
Hopcroft 2006).
In this study, we examine the effect of socioeconomic status (SES) on marriage success in the US
population of 1910, using census data. In addition,
we test an additional prediction from the theory of
biological markets ( Noë & Hammerstein 1994).
Marriage can be seen as partly involving a trade of
female fertility and nurturance for male genes,
resources and paternal investment, and, as in any
trade, prices are affected by supply and demand.
Received 29 October 2007
Accepted 16 November 2007
When men are locally abundant, women will be able
to demand a higher ‘price’ in terms of SES for
entering a marriage than they can when men are
locally scarce. Effects of this type were predicted by
Pedersen (1991) who argued that when the sex ratio
was male biased, men would have to offer greater
commitment to careers promising economic rewards,
greater fidelity and greater investment in children
than they would when the sex ratio was neutral or
female biased. However, Pedersen’s predictions have
not been tested quantitatively.
The US population of 1910 is an ideal arena in
which to test such predictions. Several decades of
male-biased migration left the relatively newly settled
western states with highly male-biased sex ratios,
while the longer settled eastern seaboard had an
equal balance of men and women (Hobbs & Stoops
2002). We thus make two basic predictions: (i) men
of high SES will be more successful in attracting a
marriage partner than men of low SES and (ii) this
effect will be moderated by the sex ratio of the state
where the individual resides. As the state sex ratio
becomes more male biased, men should have higher
relative SES in order to marry.
2. MATERIAL AND METHODS
The Integrated Public Use Microdata Series (IPUMS) sample of
1910 contains data on demographics and household composition
for the US population. It is a 1 in 250 random sample of the total
population, and is widely used by economists, historians and
sociologists (see Ruggles et al. (1997) for more details). We
calculated the operational sex ratio (OSR) for each state. Following
Lummaa et al. (1998), we define this as the ratio of males to
females between 15 and 50 years old (mean ageZ30.21 years,
s.d.Z10.15). Though some of these individuals will be married, the
number of married men and women in each state must be equal
owing to legal monogamy, and thus our OSR measure reflects the
relative availability of unmarried men and women (OSRO1
indicates an oversupply of men). The overall OSR is slightly male
biased (mean of statesZ1.02, s.d.Z0.03; mean of all individualsZ
1.01, s.d.Z0.02). OSR by the state ranges from slightly female
biased (Maine, 0.98; Connecticut, 0.99) to strongly male biased
(Montana, 1.11; Arizona, 1.10; Nevada, 1.09).
The number of men in the analysis was 21 973. SES was
measured according to the Duncan SEI score of 1950. This is a
measure based on occupational prestige (higher scores equal higher
status; see Haug 1977; Ruggles et al. 1997). Jobless men were
assigned a score of 0 (10% of the sample), which is a reasonable
inference about their relative SES in this pre-social security
population. SES scores thus range from 0 to 96 (meanZ22.38,
s.d.Z21.34). The main conclusions presented below were not
altered if jobless men were excluded rather than assigned a value of
0. Our dependent variable is the number of times a man married
(meanZ0.52, s.d.Z0.57). As men rarely married more than once
(3.5% of men married twice or more), this variable basically
measures the probability of marriage.
We examined the independent effects of the sex ratio, SES and
age on the number of times married, using general linear mixed
modelling (GLMM). The GLMM typically assumes a continuous
dependent variable, but in the present case the dependent variable
can also be taken as a count variable. Therefore, we also analysed
the data using negative binomial regression (NBR; Gardner et al.
1995), which gave similar effects as those presented below (results
not shown). However, the GLMM is more flexible than NBR in
modelling multilevel predictors, random effects and covariance
structures, so the GLMM results are presented here. Parameters in
the models are estimated by restricted maximum log likelihood. For
both the models, there was an absolute parameter, log likelihood and
Hessian convergence (Verbeke & Molenberghs 2000; SPSS 2005).
First, we constructed a model with a random intercept and then
compared these with the models with random slopes (for age and
SES) and a random intercept based on the state level. The
covariance structure of parameters was estimated in several ways
(simple, compound symmetric, autoregressive, Toeplitz and
unstructured), and based on information criteria (AIC, BIC) we
selected the best fit for the final models (Kuha 2004; see Litell
This journal is q 2007 The Royal Society
2
T. V. Pollet & D. Nettle
Driving a hard bargain
Table 1. Standardized parameter estimates from the general
linear mixed models.
25.00
s.e.
d.f.
t
p
K0.07
K0.09
0.61
0.05
K0.04
0.02
0.01
0.01
0.008
0.008
17.07
21.88
233.94
21.03
108.4
K3.55
K6.16
60.33
5.9
K4.84
0.002
!0.0001
!0.0001
!0.0001
!0.0001
K0.01
0.03
0.005
0.007
405.71
47.01
K2.73
4.01
0.007
!0.0001
et al. 2000). For the final models, we present parameter estimates,
with F-tests and correspondent t-tests for each parameter. The
F-tests are used to examine whether a variable significantly
contributes to the model, whereas the t-tests allow the examination
of individual parameter estimates.
We present just one model incorporating both the baseline and
interaction effects. Two predictions are tested: the first prediction is
that a man’s SES will influence his probability of marriage, while
controlling for age and sex ratio of the state and the second
prediction is that there will be a two-way interaction effect for
SES!sex ratio on marriage success. As the sex ratio increases, the
strength of the effect of SES on the number of times married will be
strengthened. We also control for the other interaction effects on the
number of times married, namely age!SES and age!sex ratios.
3. RESULTS
The model with the best fit is the one with random
slopes for age and SES and a random intercept at the
state level (based on BIC and AIC). It has an
unstructured covariance structure (AIC: 52194, BIC:
52250). The model with the second best fit has
random slopes for age and SES at the state level but
no random intercept (AIC: 52540, BIC: 52572) and
showed nearly identical estimates.
The F-tests show that all main effects and three
two-way interaction effects contribute to the model
(all F-tests p!0.0001, except for the F-test for the
intercept (F1,17.07Z12.63; pZ0.002) and the F-test
and for the age!SES interaction (F1,405.71Z44.45;
pZ0.007)). The standardized parameter estimates for
all the effects are presented in table 1. Our first
prediction is confirmed. High SES men are more
likely to be married than low SES men (figure 1).
Figure 1 plots the model predictions rather than the
raw data, as the model allows correcting for age (and
random effects).
Our second prediction is also confirmed. The
significant interaction between the sex ratio and the
SES means that as the state becomes more male
biased, the effect of individual SES on the number of
times married becomes stronger. To visualize this
interaction, figure 2 shows the SES of a predicted
married man relative to the SES of a predicted
unmarried man for each state against the sex ratio of
that state. Predicted married status rather than actual
status is used, as this controls for age and random
effects at the state level. Figure 2 shows that where
the sex ratios are balanced, the model predicts
married men to have just slightly higher SES than
unmarried men. As the sex ratio increases, married
men are predicted to need two or three times the SES
Biol. Lett.
20.00
SES
intercept
sex ratio
age
SES
sex ratio
!age
age!SES
sex ratio
!SES
b
15.00
10.00
5.00
unmarried
married
predicted marital status for a man
Figure 1. SES for a (predicted) married man. Bars
represent 95% CI for the mean.
8.00
SES (predicted) married men /
SES (predicted) unmarried men
interaction
model
7.00
6.00
5.00
4.00
3.00
2.00
1.00
1.00
1.05
sex ratio
1.10
Figure 2. Ratio of SES for (predicted) married men to SES
for (predicted) unmarried men plotted against sex ratio at
state level. (States where OSRO1.08 are not represented as
the model predicts no one would marry there. These are
states with a relatively small population.)
of unmarried men from the same state. The outlier
point in figure 2 represents Hawaii, which is strongly
male biased with a small population and also shows
large variance in the SES. Exclusion of Hawaii does
not alter the significance of the SES!sex ratio
interaction.
In order to illustrate the order of magnitude of the
interaction effect, we calculated the model predictions
for a 30-year-old man of low SES (1 s.d. below the
mean) and high SES (1 s.d. above the mean) in a
balanced (OSRZ1) versus male-biased state (OSRZ
1.1). In a balanced state, the model predicts that 56%
of low SES men would be married once, whereas
60% of high SES men would. In a male-biased state,
however, only 24% of low SES men would be
married once, but 46% of high SES men would.
Thus, with a sex ratio shift from 1 to 1.1, low SES
men become 2.31 times less likely to marry, whereas
high SES men are 1.31 times less likely.
Driving a hard bargain
4. DISCUSSION
Men of low SES were less likely to marry than men
with high SES. In addition, our prediction from the
theory of biological markets was met. As the sex ratio
became more male biased, the effect of SES on the
probability of marriage became stronger, such that
the relative SES required to marry became greater.
This means that the effects of a male-biased sex ratio
fell disproportionately on low SES men, whose
probability of marriage was drastically reduced.
In line with the diverse studies of both hypothetical
preferences (e.g. Buss & Barnes 1986; Buss 1989)
and actual marital outcomes (e.g. Borgerhoff Mulder
1990; Hopcroft 2006), this study thus confirms that
women prefer men of high SES, and that this
preference is influential, as the shifts induced by sex
ratio variation demonstrate. Our study confirms
Pedersen’s (1991) suggestion that changes in OSR
will have important consequences for the marriage
market. Pedersen’s predictions are much broader
than the effect described here. He suggests that
people will change many aspects of their behaviour as
a knock-on effect of the competition induced by sex
ratio fluctuations. For men, these include greater
fidelity, commitment to careers and increased investment in children. Thus, much about the varying
ethos of male and female behaviour across populations and across time could in principle be
explained with reference to the sex ratio. These
questions are ripe for future investigation, but our
study has clearly established the more limited fact
that sex ratio fluctuations in modern humans can put
one sex in the driving seat and allow them to drive a
hard bargain.
The authors wish to thank the participants of the ASAB
conference at the Newcastle University and the anonymous
reviewers for valuable feedback. The authors are grateful to
the IPUMS team (http://www.ipums.org) and the University
of Minnesota for making the data presented here publicly
available.
Borgerhoff Mulder, M. 1990 Kipsigis women’s preferences
for wealthy men: evidence for female choice in mammals. Behav. Ecol. Sociobiol. 27, 255–264. (doi:10.1007/
BF00164897)
Buss, D. M. 1989 Sex differences in human mate preferences: evolutionary hypotheses tested in 37 cultures.
Behav. Brain Sci. 12, 1–49.
Biol. Lett.
T. V. Pollet & D. Nettle
3
Buss, D. M. & Barnes, M. L. 1986 Preferences in human
mate selection. J. Pers. Soc. Psychol. 50, 559–570.
(doi:10.1037/0022-3514.50.3.559)
Buunk, B. P., Dijkstra, P., Fetchenhauer, D. & Kenrick,
D. T. 2002 Age and gender differences in mate selection
criteria for various involvement levels. Pers. Relationship
9, 271–278. (doi:10.1111/1475-6811.00018)
Gardner, W., Mulvey, E. P. & Shaw, E. C. 1995 Regression
analyses of counts and rates: Poisson, overdispersed
Poisson, and negative binomial models. Psychol. Bull.
118, 392–404. (doi:10.1037/0033-2909.118.3.392)
Haug, M. R. 1977 Measurement of social stratification.
Annu. Rev. Sociol. 3, 51–77. (doi:10.1146/annurev.so.03.
080177.000411)
Hobbs, F. & Stoops, N. 2002 Demographic trends in the 20th
century: census 2000 special reports. Washington, DC: US
Census Bureau.
Hopcroft, R. L. 2006 Sex, status and reproductive success
in the contemporary U.S. Evol. Hum. Behav. 27,
104–120. (doi:10.1016/j.evolhumbehav.2005.07.004)
Kuha, J. 2004 AIC and BIC: comparisons of assumptions
and performance. Sociol. Method Res. 33, 188–229.
(doi:10.1177/0049124103262065)
Litell, S. C., Pendergast, J. & Natarajan, R. 2000 Modelling
covariance structure in the analysis of repeated measures
data. Stat. Med. 19, 1793–1819. (doi:10.1002/1097-0258
(20000715)19:13!1793::AID-SIM482O3.0.CO;2-Q)
Low, B. S. 1990 Occupational status, land ownership, and
reproductive behavior in 19th century Sweden: Tuna
parish. Am. Anthropol. 92, 457–468. (doi:10.1525/aa.
1990.92.2.02a00130)
Lummaa, V., Merilä, J. & Kausse, A. 1998 Adaptive sex
ratio variation in pre-industrial human (Homo sapiens)
populations? Proc. R. Soc. B 265, 563–568. (doi:10.
1098/rspb.1998.0331)
Noë, R. & Hammerstein, P. 1994 Biological markets:
supply and demand determine the effect of partner
choice in cooperation, mutualism and mating. Behav.
Ecol. Sociobiol. 35, 1–11.
Pedersen, F. A. 1991 Secular trends in human sex ratios:
their influence on individual and family behavior. Hum.
Nat. Int. Bios. 2, 271–291. (doi:10.1007/BF02692189)
Pérusse, D. 1993 Cultural and reproductive success in
industrial societies: testing the relationship at the proximate and ultimate levels. Behav. Brain Sci. 16, 267–322.
Ruggles, S., Sobek, M., Fitch, C. A., Hall, P. K. &
Ronnander, C. 1997 Integrated public use microdata series:
version 2.0. Minneapolis, MN: Historical Census
Projects, University of Minnesota.
SPSS 2005 Linear mixed-effects modeling in SPSS: an
introduction to the MIXED procedure. Chicago, IL: SPSS,
Inc.
Verbeke, G. & Molenberghs, G. 2000 Linear mixed models
for longitudinal data. New York, NY: Springer.